or4k2l/Structural-Geospatial-Identification
GPS + IMU classifier for passive object identification human, car, truck, bicycle, drone. No camera, no radar
What's novel
GPS tells us where. SGI tells us what. Together: passive object identification, no camera, no radar. In Cooperation whith AI.
Code Analysis
7 files read · 3 roundsA transport-mode classifier (human, bicycle, car, truck, drone) that extracts 14 physics-based features from GPS+IMU time-series windows and classifies them with a RandomForest trained on synthetic data.
Strengths
Exceptionally honest README with real-world validation results showing where the model fails; thoughtful feature engineering with physically motivated discriminators (vib_freq_ratio, speed-normalized curvature) and a synthetic data generator that enforces domain invariants (equal road noise for car/truck).
Weaknesses
The core ML is a standard RandomForest on hand-crafted features with no novel algorithmic contribution; the 'physics' module (K-field, relativistic corrections) is purely theoretical and ~35 orders of magnitude below detectable, adding conceptual weight without functional value.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
44
Contributors
2
Files
16
Active weeks
1
Repository
Language
Jupyter Notebook
Stars
0
Forks
0
License
MIT